1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 2020 Google LLC
4 * Author: Quentin Perret <qperret@google.com>
5 */
6
7 #include <linux/kvm_host.h>
8 #include <asm/kvm_hyp.h>
9 #include <asm/kvm_mmu.h>
10 #include <asm/kvm_pgtable.h>
11
12 #include <nvhe/early_alloc.h>
13 #include <nvhe/fixed_config.h>
14 #include <nvhe/gfp.h>
15 #include <nvhe/memory.h>
16 #include <nvhe/mem_protect.h>
17 #include <nvhe/mm.h>
18 #include <nvhe/trap_handler.h>
19
20 struct hyp_pool hpool;
21 unsigned long hyp_nr_cpus;
22
23 #define hyp_percpu_size ((unsigned long)__per_cpu_end - \
24 (unsigned long)__per_cpu_start)
25
26 static void *vmemmap_base;
27 static void *hyp_pgt_base;
28 static void *host_s2_pgt_base;
29 static struct kvm_pgtable_mm_ops pkvm_pgtable_mm_ops;
30
divide_memory_pool(void * virt,unsigned long size)31 static int divide_memory_pool(void *virt, unsigned long size)
32 {
33 unsigned long vstart, vend, nr_pages;
34
35 hyp_early_alloc_init(virt, size);
36
37 hyp_vmemmap_range(__hyp_pa(virt), size, &vstart, &vend);
38 nr_pages = (vend - vstart) >> PAGE_SHIFT;
39 vmemmap_base = hyp_early_alloc_contig(nr_pages);
40 if (!vmemmap_base)
41 return -ENOMEM;
42
43 nr_pages = hyp_s1_pgtable_pages();
44 hyp_pgt_base = hyp_early_alloc_contig(nr_pages);
45 if (!hyp_pgt_base)
46 return -ENOMEM;
47
48 nr_pages = host_s2_pgtable_pages();
49 host_s2_pgt_base = hyp_early_alloc_contig(nr_pages);
50 if (!host_s2_pgt_base)
51 return -ENOMEM;
52
53 return 0;
54 }
55
recreate_hyp_mappings(phys_addr_t phys,unsigned long size,unsigned long * per_cpu_base,u32 hyp_va_bits)56 static int recreate_hyp_mappings(phys_addr_t phys, unsigned long size,
57 unsigned long *per_cpu_base,
58 u32 hyp_va_bits)
59 {
60 void *start, *end, *virt = hyp_phys_to_virt(phys);
61 unsigned long pgt_size = hyp_s1_pgtable_pages() << PAGE_SHIFT;
62 enum kvm_pgtable_prot prot;
63 int ret, i;
64
65 /* Recreate the hyp page-table using the early page allocator */
66 hyp_early_alloc_init(hyp_pgt_base, pgt_size);
67 ret = kvm_pgtable_hyp_init(&pkvm_pgtable, hyp_va_bits,
68 &hyp_early_alloc_mm_ops);
69 if (ret)
70 return ret;
71
72 ret = hyp_create_idmap(hyp_va_bits);
73 if (ret)
74 return ret;
75
76 ret = hyp_map_vectors();
77 if (ret)
78 return ret;
79
80 ret = hyp_back_vmemmap(phys, size, hyp_virt_to_phys(vmemmap_base));
81 if (ret)
82 return ret;
83
84 ret = pkvm_create_mappings(__hyp_text_start, __hyp_text_end, PAGE_HYP_EXEC);
85 if (ret)
86 return ret;
87
88 ret = pkvm_create_mappings(__hyp_rodata_start, __hyp_rodata_end, PAGE_HYP_RO);
89 if (ret)
90 return ret;
91
92 ret = pkvm_create_mappings(__hyp_bss_start, __hyp_bss_end, PAGE_HYP);
93 if (ret)
94 return ret;
95
96 ret = pkvm_create_mappings(virt, virt + size, PAGE_HYP);
97 if (ret)
98 return ret;
99
100 for (i = 0; i < hyp_nr_cpus; i++) {
101 start = (void *)kern_hyp_va(per_cpu_base[i]);
102 end = start + PAGE_ALIGN(hyp_percpu_size);
103 ret = pkvm_create_mappings(start, end, PAGE_HYP);
104 if (ret)
105 return ret;
106
107 end = (void *)per_cpu_ptr(&kvm_init_params, i)->stack_hyp_va;
108 start = end - PAGE_SIZE;
109 ret = pkvm_create_mappings(start, end, PAGE_HYP);
110 if (ret)
111 return ret;
112 }
113
114 /*
115 * Map the host's .bss and .rodata sections RO in the hypervisor, but
116 * transfer the ownership from the host to the hypervisor itself to
117 * make sure it can't be donated or shared with another entity.
118 *
119 * The ownership transition requires matching changes in the host
120 * stage-2. This will be done later (see finalize_host_mappings()) once
121 * the hyp_vmemmap is addressable.
122 */
123 prot = pkvm_mkstate(PAGE_HYP_RO, PKVM_PAGE_SHARED_OWNED);
124 ret = pkvm_create_mappings(__start_rodata, __end_rodata, prot);
125 if (ret)
126 return ret;
127
128 ret = pkvm_create_mappings(__hyp_bss_end, __bss_stop, prot);
129 if (ret)
130 return ret;
131
132 return 0;
133 }
134
update_nvhe_init_params(void)135 static void update_nvhe_init_params(void)
136 {
137 struct kvm_nvhe_init_params *params;
138 unsigned long i;
139
140 for (i = 0; i < hyp_nr_cpus; i++) {
141 params = per_cpu_ptr(&kvm_init_params, i);
142 params->pgd_pa = __hyp_pa(pkvm_pgtable.pgd);
143 dcache_clean_inval_poc((unsigned long)params,
144 (unsigned long)params + sizeof(*params));
145 }
146 }
147
hyp_zalloc_hyp_page(void * arg)148 static void *hyp_zalloc_hyp_page(void *arg)
149 {
150 return hyp_alloc_pages(&hpool, 0);
151 }
152
hpool_get_page(void * addr)153 static void hpool_get_page(void *addr)
154 {
155 hyp_get_page(&hpool, addr);
156 }
157
hpool_put_page(void * addr)158 static void hpool_put_page(void *addr)
159 {
160 hyp_put_page(&hpool, addr);
161 }
162
finalize_host_mappings_walker(u64 addr,u64 end,u32 level,kvm_pte_t * ptep,enum kvm_pgtable_walk_flags flag,void * const arg)163 static int finalize_host_mappings_walker(u64 addr, u64 end, u32 level,
164 kvm_pte_t *ptep,
165 enum kvm_pgtable_walk_flags flag,
166 void * const arg)
167 {
168 enum kvm_pgtable_prot prot;
169 enum pkvm_page_state state;
170 kvm_pte_t pte = *ptep;
171 phys_addr_t phys;
172
173 if (!kvm_pte_valid(pte))
174 return 0;
175
176 if (level != (KVM_PGTABLE_MAX_LEVELS - 1))
177 return -EINVAL;
178
179 phys = kvm_pte_to_phys(pte);
180 if (!addr_is_memory(phys))
181 return -EINVAL;
182
183 /*
184 * Adjust the host stage-2 mappings to match the ownership attributes
185 * configured in the hypervisor stage-1.
186 */
187 state = pkvm_getstate(kvm_pgtable_hyp_pte_prot(pte));
188 switch (state) {
189 case PKVM_PAGE_OWNED:
190 return host_stage2_set_owner_locked(phys, PAGE_SIZE, pkvm_hyp_id);
191 case PKVM_PAGE_SHARED_OWNED:
192 prot = pkvm_mkstate(PKVM_HOST_MEM_PROT, PKVM_PAGE_SHARED_BORROWED);
193 break;
194 case PKVM_PAGE_SHARED_BORROWED:
195 prot = pkvm_mkstate(PKVM_HOST_MEM_PROT, PKVM_PAGE_SHARED_OWNED);
196 break;
197 default:
198 return -EINVAL;
199 }
200
201 return host_stage2_idmap_locked(phys, PAGE_SIZE, prot);
202 }
203
finalize_host_mappings(void)204 static int finalize_host_mappings(void)
205 {
206 struct kvm_pgtable_walker walker = {
207 .cb = finalize_host_mappings_walker,
208 .flags = KVM_PGTABLE_WALK_LEAF,
209 };
210 int i, ret;
211
212 for (i = 0; i < hyp_memblock_nr; i++) {
213 struct memblock_region *reg = &hyp_memory[i];
214 u64 start = (u64)hyp_phys_to_virt(reg->base);
215
216 ret = kvm_pgtable_walk(&pkvm_pgtable, start, reg->size, &walker);
217 if (ret)
218 return ret;
219 }
220
221 return 0;
222 }
223
__pkvm_init_finalise(void)224 void __noreturn __pkvm_init_finalise(void)
225 {
226 struct kvm_host_data *host_data = this_cpu_ptr(&kvm_host_data);
227 struct kvm_cpu_context *host_ctxt = &host_data->host_ctxt;
228 unsigned long nr_pages, reserved_pages, pfn;
229 int ret;
230
231 /* Now that the vmemmap is backed, install the full-fledged allocator */
232 pfn = hyp_virt_to_pfn(hyp_pgt_base);
233 nr_pages = hyp_s1_pgtable_pages();
234 reserved_pages = hyp_early_alloc_nr_used_pages();
235 ret = hyp_pool_init(&hpool, pfn, nr_pages, reserved_pages);
236 if (ret)
237 goto out;
238
239 ret = kvm_host_prepare_stage2(host_s2_pgt_base);
240 if (ret)
241 goto out;
242
243 ret = finalize_host_mappings();
244 if (ret)
245 goto out;
246
247 pkvm_pgtable_mm_ops = (struct kvm_pgtable_mm_ops) {
248 .zalloc_page = hyp_zalloc_hyp_page,
249 .phys_to_virt = hyp_phys_to_virt,
250 .virt_to_phys = hyp_virt_to_phys,
251 .get_page = hpool_get_page,
252 .put_page = hpool_put_page,
253 };
254 pkvm_pgtable.mm_ops = &pkvm_pgtable_mm_ops;
255
256 out:
257 /*
258 * We tail-called to here from handle___pkvm_init() and will not return,
259 * so make sure to propagate the return value to the host.
260 */
261 cpu_reg(host_ctxt, 1) = ret;
262
263 __host_enter(host_ctxt);
264 }
265
__pkvm_init(phys_addr_t phys,unsigned long size,unsigned long nr_cpus,unsigned long * per_cpu_base,u32 hyp_va_bits)266 int __pkvm_init(phys_addr_t phys, unsigned long size, unsigned long nr_cpus,
267 unsigned long *per_cpu_base, u32 hyp_va_bits)
268 {
269 struct kvm_nvhe_init_params *params;
270 void *virt = hyp_phys_to_virt(phys);
271 void (*fn)(phys_addr_t params_pa, void *finalize_fn_va);
272 int ret;
273
274 BUG_ON(kvm_check_pvm_sysreg_table());
275
276 if (!PAGE_ALIGNED(phys) || !PAGE_ALIGNED(size))
277 return -EINVAL;
278
279 hyp_spin_lock_init(&pkvm_pgd_lock);
280 hyp_nr_cpus = nr_cpus;
281
282 ret = divide_memory_pool(virt, size);
283 if (ret)
284 return ret;
285
286 ret = recreate_hyp_mappings(phys, size, per_cpu_base, hyp_va_bits);
287 if (ret)
288 return ret;
289
290 update_nvhe_init_params();
291
292 /* Jump in the idmap page to switch to the new page-tables */
293 params = this_cpu_ptr(&kvm_init_params);
294 fn = (typeof(fn))__hyp_pa(__pkvm_init_switch_pgd);
295 fn(__hyp_pa(params), __pkvm_init_finalise);
296
297 unreachable();
298 }
299